vm: use error wrapping to detect ssh connection errors

This is a much cleaner logic than string matching.

Author: a-nogikh

vm/qemu/qemu.go

@@ -7,6 +7,7 @@ import (
 	"bytes"
 	"context"
 	"encoding/json"
+	"errors"
 	"fmt"
 	"io"
 	"net"
@@ -348,10 +349,8 @@ func (pool *Pool) Create(ctx context.Context, workdir string, index int) (vmimpl
 		if err == nil {
 			return inst, nil
 		}
-		if strings.Contains(err.Error(), "can't ssh into the instance") {
-			// If there was a boot error, immediately return it.
-			// In this case, the string search below also matches against boot time output,
-			// and e.g. "Device or resource busy" is quite often in there.
+		if errors.Is(err, vmimpl.ErrCantSSH) {
+			// It is most likely a boot crash, just return the error as is.
 			return nil, err
 		}
 		// Older qemu prints "could", newer -- "Could".

vm/vmimpl/util.go

@@ -58,13 +58,15 @@ func WaitForSSH(timeout time.Duration, opts SSHOptions, OS string, stop chan err
 		}
 		if time.Since(startTime) > timeout {
 			return &osutil.VerboseError{
-				Err:    fmt.Errorf("can't ssh into the instance"),
+				Err:    ErrCantSSH,
 				Output: []byte(err.Error()),
 			}
 		}
 	}
 }
 
+var ErrCantSSH = fmt.Errorf("can't ssh into the instance")
+
 func SSHArgs(debug bool, sshKey string, port int, systemSSHCfg bool) []string {
 	return sshArgs(debug, sshKey, "-p", port, 0, systemSSHCfg)
 }